The 2016 U.S. Presidential Candidates and How People Tweeted About Them

The 2016 election provided more language and polling data than any previous election. In addition, the election spurred a new level of social media coverage. The current study analyzed the language of Donald Trump and Hillary Clinton from the debates as well as the tweets of millions of people during the fall presidential campaign. In addition, aggregated polling data allowed for a comparison of daily election-relevant language from Twitter and fluctuations in voter preference. Overall, Trump’s debate language was low in analytic/formal thinking and high in negative emotional tone and authenticity. Clinton was high in analytic and positive emotions, low in authenticity. The analysis of almost 10 million tweets revealed that Trump-relevant tweets were generally more positive than Clinton-related tweets. Most important were lag analyses that predicted polling numbers a week later from tweets. In general, when Clinton-related tweets were more analytic, her subsequent poll numbers dropped. Similarly, positive emotion language in the Clinton-related tweets predicted lower poll numbers a week later. Conversely, Trump-related tweets that were high in positive emotion and in analytic thinking predicted higher subsequent polling. In other words, when Twitter language about the candidates was used in ways inconsistent with the candidates themselves, their poll numbers went up. We propose two possible explanations for these findings: the projection hypothesis, a desire to seek qualities the candidates are missing, and the participant hypothesis, a shift in who is participating in the Twitter conversation over the course of the campaigns.